From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 29533 invoked by alias); 16 Apr 2015 19:13:38 -0000 Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Received: (qmail 29515 invoked by uid 89); 16 Apr 2015 19:13:37 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-1.3 required=5.0 tests=AWL,BAYES_00,FREEMAIL_FROM,RCVD_IN_DNSWL_LOW,SPF_PASS autolearn=ham version=3.3.2 X-Spam-User: qpsmtpd, 2 recipients X-HELO: mail-wg0-f54.google.com Received: from mail-wg0-f54.google.com (HELO mail-wg0-f54.google.com) (74.125.82.54) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with (AES128-GCM-SHA256 encrypted) ESMTPS; Thu, 16 Apr 2015 19:13:34 +0000 Received: by wgyo15 with SMTP id o15so91458849wgy.2; Thu, 16 Apr 2015 12:13:31 -0700 (PDT) MIME-Version: 1.0 X-Received: by 10.181.13.113 with SMTP id ex17mr10846257wid.12.1429211611772; Thu, 16 Apr 2015 12:13:31 -0700 (PDT) Received: by 10.180.82.133 with HTTP; Thu, 16 Apr 2015 12:13:31 -0700 (PDT) In-Reply-To: <20150414190054.473a9bbb@gmx.de> References: <20150226181717.480e282c@vepi2> <551006FF.1080704@sfr.fr> <20150323134357.6af740d1@vepi2> <20150324180620.3c72960e@vepi2> <20150409143709.6d33aa8c@vepi2> <20150414190054.473a9bbb@gmx.de> Date: Thu, 16 Apr 2015 19:13:00 -0000 Message-ID: Subject: Re: [Patch, Fortran, pr60322, addendum] was: [Patch 1/2, Fortran, pr60322] [OOP] Incorrect bounds on polymorphic dummy array From: Paul Richard Thomas To: Andre Vehreschild Cc: Mikael Morin , GCC-Fortran-ML , GCC-Patches-ML , Antony Lewis , Dominique Dhumieres Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable X-SW-Source: 2015-04/txt/msg00833.txt.bz2 Hi Andre, The delta patch is OK for trunk and eventual backport to 5.2. Thanks for all the hard work Paul On 14 April 2015 at 19:00, Andre Vehreschild wrote: > Hi all, > > during further testing of a big Fortran software I encounter two bugs with > class arrays, that are somehow connected to pr60322. I therefore propose = an > extended patch for pr60322. Because Paul has already reviewed most the ex= tended > patch, I give you two patches: > > 1. a full patch, fixing all the issues connected to pr60322, and > 2. a delta patch to get from the reviewed patch to the latest version. > > With the second patch I hope to get a faster review, because it is > significantly shorter. > > Now what was the issue? To be precise there were two issues: > > i. a pointer to a class array (CLASS_DATA(sym).attr.class_pointer =3D=3D = 1) was > dereferenced, which lead to an ICE (the patch for this in the delta is ch= unk 5 > in gfc_conv_expr_descriptor, and > > ii. (and this was a severe brain cracker) in chains of references consist= ing of > more then one class-(array)-ref always the _vptr of the first symbol was = taken > and not the _vptr of the currently dereferenced class object. This occurr= ed > when fortran code similiar to this was executed: > > type innerT > integer, allocatable :: arr(:) > end type > > type T > class(innerT) :: mat(:,:) > end type > > class(T) :: o > > allocate(o%mat(2,2)) > allocate(o%mat(:,:)%arr(10)) ! This is obviously pseudo code, > ! but I think you get what is meant. > > o%mat(1,1)%arr(1) =3D 1 > > In the last line the address to get to arr(1) was computed using the > _vptr->size of o and not of o%mat(1,1). To fix this gfc_component_ref () = now > computes the class' _vptr-ref whenever it does a _data-ref (chunk 1 of > trans-expr.c in the delta patch). The _vptr-ref is stored in gfc_se, wher= e I > added the new member class_vptr. The gfc_se->class_vptr is then used in > array-refs (chunk 2 of trans.c) to get the size of the array elements of = the > correct level. > > The other chunks of the delta patch are: > - parameter passing fixes, and > - documentation fixes as requested for the version 5 of the pr60322 patch. > > I hope this helps in getting the patch reviewed quickly. > > Bootstraps and regtests ok on x86_64-linux-gnu/F21. > > Ok for trunk -> 6.0? > Ok, for backport to 5.2, once available? > > Note, the patches may apply with shifts, as I forgot to update before tak= ing > the diffs. > > Regards, > Andre > > On Thu, 9 Apr 2015 14:37:09 +0200 > Andre Vehreschild wrote: > >> Hi Paul, hi all, >> >> Paul, thanks for the review. Answers to your questions are inline below: >> >> On Sun, 5 Apr 2015 11:13:05 +0200 >> Paul Richard Thomas wrote: >> >> > + /* The dummy is returned for pointer, allocatable or assumed rank a= rrays. >> > + The check for pointerness needs to be repeated here (it is done = in >> > + IS_CLASS_ARRAY (), too), because for class arrays that are point= ers, >> > as >> > + is the one of the sym, which is incorrect here. */ >> > >> > What does this mean, please? >> >> The first sentence is about regular arrays and should be unchanged from = the >> original source. Then I have to check for class (arrays) that are pointe= rs, >> i.e., independent of whether the sym is a class array or a regular point= er to >> a class object. (The latter shouldn't make it into the routine anyway.) >> IS_CLASS_ARRAY () returns false for too many reasons to be of use here. = I have >> to apologize and confess that the comment was a mere note to myself to n= ot >> return to use is_classarray in the if below. Let me rephrase the comment= to >> be: >> >> /* The dummy is returned for pointer, allocatable or assumed rank arrays. >> For class arrays the information if sym is an allocatable or pointer >> object needs to be checked explicitly (IS_CLASS_ARRAY can be false for >> too many reasons to be of use here). */ >> >> > + /* Returning the descriptor for dummy class arrays is hazardous, >> > because >> > + some caller is expecting an expression to apply the component re= fs to. >> > + Therefore the descriptor is only created and stored in >> > + sym->backend_decl's GFC_DECL_SAVED_DESCRIPTOR. The caller is th= en >> > + responsible to extract it from there, when the descriptor is >> > + desired. */ >> > + if (IS_CLASS_ARRAY (sym) >> > + && (!DECL_LANG_SPECIFIC (sym->backend_decl) >> > + || !GFC_DECL_SAVED_DESCRIPTOR (sym->backend_decl))) >> > + { >> > + decl =3D gfc_build_dummy_array_decl (sym, sym->backend_decl); >> > + /* Prevent the dummy from being detected as unused if it is cop= ied. >> > */ >> > + if (sym->backend_decl !=3D NULL && decl !=3D sym->backend_decl) >> > + DECL_ARTIFICIAL (sym->backend_decl) =3D 1; >> > + sym->backend_decl =3D decl; >> > + } >> > >> > The comments, such as the above are often going well beyond column 72, >> > into the 80's. I know that much of the existing code violates this >> > style requirement but there is no need to do so if clarity is not >> > reduced thereby. >> >> Er, the document at >> >> https://gcc.gnu.org/codingconventions.html#C_Formatting >> >> says that line length is 80, or is there another convention, that I am n= ot >> aware of? >> >> > In trans-stmt.c s/standart/standard/ >> >> Fixed. >> >> > Don't forget to put the PR numbers in the ChangeLogs. >> >> I won't anymore, already got told off :-) >> >> > For this submission, I would have appreciated some a description of >> > what each chunk in the patch is doing, just because there is so much >> > of it. I suppose that it was good for my imortal soul to sort it out >> > for myself but it took a little while :-) >> >> I initially tried to split the submission in two parts to make it more >> manageable. One part with the brain-dead substitutions of as and array_a= ttr >> and one with the new code. Albeit I failed to get the brain-dead part ri= ght >> and made some mistakes there already, which Mikael pointed out. I theref= ore >> went for the big submission. >> >> Now doing a description of what each "chunk" does is quite tedious. I re= ally >> would like to spend my time more productive. Would you be satisfied, whe= n I >> write a story about the patch, referring to some parts more explicitly, = like >> >> "Chunk 4 of file trans-stmt.c is the heart of the patch and does this and >> that. The remaining chunks are more or less putting the data together." >> >> (This is not correct for this patch of course. Just an example.) More >> elaborate of course, but just to give an idea. >> >> Thanks again. I will commit as soon as 5.2/6.0 commit window is open. >> >> Regards, >> Andre >> >> > >> > Cheers and many thanks for the patch. >> > >> > Paul >> > >> > On 27 March 2015 at 13:48, Paul Richard Thomas >> > wrote: >> > > Dear Andre, >> > > >> > > I am in the UK as of last night. Before leaving, I bootstrapped and >> > > regtested your patch and all was well. I must drive to Cambridge this >> > > afternoon to see my mother and will try to get to it either this >> > > evening or tomorrow morning. There is so much of it and it touches >> > > many places; so I must give it a very careful looking over before >> > > giving the green light. Bear with me please. >> > > >> > > Great work though! >> > > >> > > Paul >> > > >> > > On 24 March 2015 at 18:06, Andre Vehreschild wrote: >> > >> Hi all, >> > >> >> > >> I have worked on the comments Mikael gave me. I am now checking for >> > >> class_pointer in the way he pointed out. >> > >> >> > >> Furthermore did I *join the two parts* of the patch into this one, >> > >> because keeping both in sync was no benefit but only tedious and di= d not >> > >> prove to be reviewed faster. >> > >> >> > >> Paul, Dominique: I have addressed the LOC issue that came up lately= . Or >> > >> rather the patch addressed it already. I feel like this is not test= ed >> > >> very well, not the loc() call nor the sizeof() call as given in the= 57305 >> > >> second's download. Unfortunately, is that download not runable. I w= ould >> > >> love to see a test similar to that download, but couldn't come up w= ith >> > >> one, that satisfied me. Given that the patch's review will last some >> > >> days, I still have enough time to come up with something beautiful = which >> > >> I will add then. >> > >> >> > >> Bootstraps and regtests ok on x86_64-linux-gnu/F20. >> > >> >> > >> Regards, >> > >> Andre >> > >> >> > >> >> > >> On Tue, 24 Mar 2015 11:13:27 +0100 >> > >> Paul Richard Thomas wrote: >> > >> >> > >>> Dear Andre, >> > >>> >> > >>> Dominique pointed out to me that the 'loc' patch causes a ICE in t= he >> > >>> testsuite. It seems that 'loc' should provide the address of the c= lass >> > >>> container in some places and the address of the data in others. I = will >> > >>> put my thinking cap on tonight :-) >> > >>> >> > >>> Cheers >> > >>> >> > >>> Paul >> > >>> >> > >>> On 23 March 2015 at 13:43, Andre Vehreschild wrote: >> > >>> > Hi Mikael, >> > >>> > >> > >>> > thanks for looking at the patch. Please note, that Paul has sent= an >> > >>> > addendum to the patches for 60322, which I deliberately have att= ached. >> > >>> > >> > >>> >> 26/02/2015 18:17, Andre Vehreschild a =E9crit : >> > >>> >> > This first patch is only preparatory and does not change any = of the >> > >>> >> > semantics of gfortran at all. >> > >>> >> Sure? >> > >>> > >> > >>> > With the counterexample you found below, this of course is a wro= ng >> > >>> > statement. >> > >>> > >> > >>> >> > diff --git a/gcc/fortran/expr.c b/gcc/fortran/expr.c >> > >>> >> > index ab6f7a5..d28cf77 100644 >> > >>> >> > --- a/gcc/fortran/expr.c >> > >>> >> > +++ b/gcc/fortran/expr.c >> > >>> >> > @@ -4059,10 +4060,10 @@ gfc_lval_expr_from_sym (gfc_symbol *s= ym) >> > >>> >> > lval->symtree =3D gfc_find_symtree (sym->ns->sym_root, sym= ->name); >> > >>> >> > >> > >>> >> > /* It will always be a full array. */ >> > >>> >> > - lval->rank =3D sym->as ? sym->as->rank : 0; >> > >>> >> > + as =3D sym->as; >> > >>> >> > + lval->rank =3D as ? as->rank : 0; >> > >>> >> > if (lval->rank) >> > >>> >> > - gfc_add_full_array_ref (lval, sym->ts.type =3D=3D BT_CLA= SS ? >> > >>> >> > - CLASS_DATA (sym)->as : sym->as); >> > >>> >> > + gfc_add_full_array_ref (lval, as); >> > >>> >> >> > >>> >> This is a change of semantics. Or do you know that sym->ts.typ= e !=3D >> > >>> >> BT_CLASS? >> > >>> > >> > >>> > You are completely right. I have made a mistake here. I have to = tell >> > >>> > the truth, I never ran a regtest with only part 1 of the patches >> > >>> > applied. The second part of the patch will correct this, by sett= ing >> > >>> > the variable as depending on whether type =3D=3D BT_CLASS or not= . Sorry >> > >>> > for the mistake. >> > >>> > >> > >>> >> > diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-dec= l.c >> > >>> >> > index 3664824..e571a17 100644 >> > >>> >> > --- a/gcc/fortran/trans-decl.c >> > >>> >> > +++ b/gcc/fortran/trans-decl.c >> > >>> >> > @@ -1013,16 +1017,24 @@ gfc_build_dummy_array_decl (gfc_symbo= l * >> > >>> >> > sym, tree dummy) tree decl; >> > >>> >> > tree type; >> > >>> >> > gfc_array_spec *as; >> > >>> >> > + symbol_attribute *array_attr; >> > >>> >> > char *name; >> > >>> >> > gfc_packed packed; >> > >>> >> > int n; >> > >>> >> > bool known_size; >> > >>> >> > >> > >>> >> > - if (sym->attr.pointer || sym->attr.allocatable >> > >>> >> > - || (sym->as && sym->as->type =3D=3D AS_ASSUMED_RANK)) >> > >>> >> > + /* Use the array as and attr. */ >> > >>> >> > + as =3D sym->as; >> > >>> >> > + array_attr =3D &sym->attr; >> > >>> >> > + >> > >>> >> > + /* The pointer attribute is always set on a _data componen= t, >> > >>> >> > therefore check >> > >>> >> > + the sym's attribute only. */ >> > >>> >> > + if (sym->attr.pointer || array_attr->allocatable >> > >>> >> > + || (as && as->type =3D=3D AS_ASSUMED_RANK)) >> > >>> >> > return dummy; >> > >>> >> > >> > >>> >> Any reason to sometimes use array_attr, sometimes not, like her= e? >> > >>> >> By the way, the comment is misleading: for classes, there is the >> > >>> >> class_pointer attribute (and it is a pain, I know). >> > >>> > >> > >>> > Yes, and a good one. Array_attr is sometimes sym->attr and somet= imes >> > >>> > CLASS_DATA(sym)->attr aka sym->ts.u.derived->components->attr. I= n the >> > >>> > later case .pointer is always set to 1 in the _data component's = attr. >> > >>> > I.e., the above if, would always yield true for a class_array, w= hich >> > >>> > is not intended, but rather destructive. I know about the >> > >>> > class_pointer attribute, but I figured, that it is not relevant = here. >> > >>> > Any idea how to formulate the comment better, to reflect what I = just >> > >>> > explained? >> > >>> > >> > >>> > Regards, >> > >>> > Andre >> > >>> > -- >> > >>> > Andre Vehreschild * Email: vehre ad gmx dot de >> > >>> > >> > >>> > >> > >>> > ---------- Forwarded message ---------- >> > >>> > From: Paul Richard Thomas >> > >>> > To: Andre Vehreschild , Dominique Dhumieres >> > >>> > Cc: >> > >>> > Date: Sun, 22 Mar 2015 21:20:20 +0100 >> > >>> > Subject: Bug in intrinsic LOC for scalar class objects >> > >>> > Dear Andre and Dominique, >> > >>> > >> > >>> > I have found that LOC is returning the address of the class cont= ainer >> > >>> > rather than the _data component for class scalars. See the source >> > >>> > below, which you will recognise! A fix is attached. >> > >>> > >> > >>> > Note that the scalar allocate fails with MOLD=3D and so I substi= tuted >> > >>> > SOURCE=3D. >> > >>> > >> > >>> > Cheers >> > >>> > >> > >>> > Paul >> > >>> > >> > >>> > class(*), allocatable :: a(:), e ! Change 'e' to an array and >> > >>> > second memcpy works correctly >> > >>> > ! Problem is with loc(e), w= hich >> > >>> > returns the address of the >> > >>> > ! class container. >> > >>> > allocate (e, source =3D 99.0) >> > >>> > allocate (a(2), source =3D [1.0, 2.0]) >> > >>> > call add_element_poly (a,e) >> > >>> > select type (a) >> > >>> > type is (real) >> > >>> > print *, a >> > >>> > end select >> > >>> > >> > >>> > contains >> > >>> > >> > >>> > subroutine add_element_poly(a,e) >> > >>> > use iso_c_binding >> > >>> > class(*),allocatable,intent(inout),target :: a(:) >> > >>> > class(*),intent(in),target :: e >> > >>> > class(*),allocatable,target :: tmp(:) >> > >>> > type(c_ptr) :: dummy >> > >>> > >> > >>> > interface >> > >>> > function memcpy(dest,src,n) bind(C,name=3D"memcpy") resu= lt(res) >> > >>> > import >> > >>> > type(c_ptr) :: res >> > >>> > integer(c_intptr_t),value :: dest >> > >>> > integer(c_intptr_t),value :: src >> > >>> > integer(c_size_t),value :: n >> > >>> > end function >> > >>> > end interface >> > >>> > >> > >>> > if (.not.allocated(a)) then >> > >>> > allocate(a(1), source=3De) >> > >>> > else >> > >>> > allocate(tmp(size(a)),source=3Da) >> > >>> > deallocate(a) >> > >>> > allocate(a(size(tmp)+1),source=3De) ! mold gives a segfa= ult >> > >>> > dummy =3D memcpy(loc(a(1)),loc(tmp),sizeof(tmp)) >> > >>> > dummy =3D memcpy(loc(a(size(tmp)+1)),loc(e),sizeof(e)) >> > >>> > end if >> > >>> > end subroutine >> > >>> > end >> > >>> > >> > >>> >> > >>> >> > >>> >> > >> >> > >> >> > >> -- >> > >> Andre Vehreschild * Email: vehre ad gmx dot de >> > > >> > > >> > > >> > > -- >> > > Outside of a dog, a book is a man's best friend. Inside of a dog it's >> > > too dark to read. >> > > >> > > Groucho Marx >> > >> > >> > >> >> > > > -- > Andre Vehreschild * Email: vehre ad gmx dot de --=20 Outside of a dog, a book is a man's best friend. Inside of a dog it's too dark to read. Groucho Marx